Anneal: upstream capability opportunities, simplification bundles, and alternative architectures

Open
#3,736 3 comments 0 reactions 0 assignees View on GitHub

Maintainers usually reply within 1 day

Nobody has claimed this yet.

Assessment

Difficulty
1/5
Estimated time
Under an hour
Newbie friendliness
55/100
Issue type
Documentation
Clarity
Clearly specified
Activity status
Active
Tech stack
rust
Domain
documentation

Research direction

Read the overview and the three linked parts of this issue: Parts I–III contain opportunities U01–U39, bundles B1–B12, alternatives, and qualifications. The issue is an exploratory catalog preserving a prior review, not an implementation request. Done means understanding or using the documented analysis; it does not ask for a code change.

Written by the indexing model from the issue text.

Description

How much simpler could Anneal be with stronger upstream contracts?

Status: exploratory opportunity catalog, not an adopted architecture, implementation roadmap, feasibility estimate, or claim that the proposed upstream capabilities exist. This issue preserves the full counterfactual review of #3735: 39 opportunities (U01–U39), 12 coordinated change packages (B1–B12), independence and overlap analysis, five larger architecture alternatives, six genuine incompatibilities, irreducible obligations, and four possible programs of work. The identifiers retain their meanings from the originating review.

The question is not merely how to optimize the existing adapters. It is: what could Anneal delete, simplify, or generalize if Cargo/rustc, Charon, Aeneas, Lean/Lake, editors, and execution infrastructure supplied stronger behavior or guarantees? Some improvements require several changes together. Others can be delivered independently. Some are alternatives whose benefits should not be counted twice.

Publication of this catalog does not adopt any option, close an implementation gate, authorize upstream changes, or supersede #3735. It records possibilities so that later investigation can choose among them.

Commission and preservation boundary

The user requested a deep review of simplifications or improvements blocked by external systems, allowing a counterfactual world in which those systems could gain arbitrary new behavior or properties. The explicit goals were to identify coordinated groups of changes, independently useful and non-overlapping groups, and mutually exclusive improvements or upstream changes. The user then directed: “File these all in a new issue.” This issue preserves the resulting review rather than reducing it to a shortlist.

The thought experiment retains Anneal's correctness, trust, source-authority, ordinary-user, and workflow obligations. It does not make an incorrect program correct, grant an agent permission to edit source, or redefine a failed request as a success. A proposed capability's hypothetical contract is an assumption for assessing its benefit, not evidence that an implementation already meets it.

Main conclusion

The design could become substantially smaller—and some parts could disappear entirely—if external contracts change rather than merely tool speed or diagnostics. The largest opportunities are to make compilation inputs explicit, make ordinary proof editing genuinely independent of program compilation, make proof contexts explicit immutable values, and make verification evidence independently consumable.

These improvements are not one indivisible project. Exact Lean snapshot queries, read-only dependency packages, compiler-authenticated annotation metadata, and editor-side transactions can each improve Anneal without waiting for the others. Conversely, a compiler-invisible proof channel is much more useful when combined with fixed proof targets and an operation that can modify only the corresponding owned proof.

Kind of improvement What counts
Remove a dependency or representation Proof text cannot affect compilation; generated models need not become temporary source workspaces; cold reconstruction is cheap enough to remove persistent caches. This can genuinely reduce total complexity.
Move responsibility to the system that owns the state Lean identifies the actual context it queried; an editor atomically validates and applies its own edits. Anneal's reconstruction machinery disappears, although the responsibility remains upstream.
Replace unchecked reasoning with checked evidence A compiler or translator supplies a checkable certificate. This can shrink unchecked trust and acceptance machinery, but the correspondence still needs a definition and a checker.

Merely putting the current orchestration behind an upstream verify() API would mostly relocate complexity. The more interesting changes make invalid combinations unrepresentable, eliminate a transformation, or establish a reusable theorem once.

Complete document and navigation

The full review is published in the three linked parts below. Each part was read back from GitHub, including its ending and authorship footer. The opening issue body is an overview, not a substitute for those parts:

Part Contents
I — Compilation, translation, and explicit proof contexts U01–U18, preserving each proposed external change, Anneal simplification, independent value or required companions, and limiting counterexample.
II — Artifacts, evidence, source authority, and ordinary use U19–U39, including stateless execution, certificate-based verification, editor transactions, stronger automation, and evidence-first proof workflows.
III — Bundles, independence, alternatives, and irreducible obligations B1–B12; worked bundle arguments; semantic/deployment/implementation independence; overlapping benefits; architecture alternatives A–E; genuine incompatibilities; what must remain; and four possible programs of work.

U and B numbers are local to this opportunity catalog; they are not #3735's D/F/T/G identifiers and do not confer completion status. All 39 opportunities remain individually described rather than being replaced by the bundle table. All counts describe published document contents, not completed implementation work or executed tests.

Governing baseline and publication-time update

The review analyzed #3735 as written against the then-current #3719 draft at 9b2a98b6be82a9f05956bd15ff6a13907630e9e7. The reference evidence used in that discussion was principally archived at reference@ba4556c30835b56d108b0a6854a760bccf8438c5, with older report snapshots identified by their links. An archive commit identifies report text, not the toolchain examined.

At filing, #3719 had advanced to 6039f3c23d42b1606610d7952b5d3ecde863a6dd and remained open and unmerged. The new draft was read before publication. Anneal's principles remain authoritative; their current main contents were rechecked and retained blob d5339a95254eae14ac201139d07d9d36d48a19fb.

The following qualifications apply throughout the catalog without removing or renumbering its opportunities:

  • Rust-source UB-freedom and preservation to compiled behavior are distinct obligations. Defined machine-code behavior alone does not prove the Rust source UB-free. U25/U26 and B7 must preserve both, including the interpretation of each guarantee's requirements and the relevant physical-system assumptions.
  • Evidence must establish the requested baseline and claim. An upstream certificate or policy checker does not provide a general test for semantic non-circularity, non-vacuity, consistency, or specification adequacy. In particular, a proof graph or an “external” label is not evidence that an assumption is independent of the conclusion. U09/U15/U25/U28/U36 do not promise such an oracle.
  • Library verification requires final contextual safety for the permitted clients. An abstract-replacement proof is one possible route, not a required architecture or a sufficient theorem by itself. U27/B7/B8 preserve this stronger final obligation. The precise permitted type/library invariants and possible unsafe-API relaxation remain unresolved lower-level semantics.
  • Trust accounting includes policy checking, evidence checking, and claim/TCB identification. Dependencies need to be complete enough for the accepted justification, not logically minimal. Policy compliance does not establish that authorized assumptions are jointly valid, consistent, or useful. U28's proposed structured trust data retains these limits.

These are publication-time compatibility qualifications, not a claim that #3735 has itself been revised or that every proposed implementation has been revalidated against the new draft. No existing issue or repository file is changed by this publication.

How to interpret the catalog

“Missing” means missing from a sufficiently strong contract established by the design, not necessarily absent from every current implementation. Existing ingredients can be starting points. For example, Lean exposes internal command snapshots and server-owned RPC objects; the examined newer Aeneas source adds translation metadata. Those facts do not alone establish the complete proposed contracts. Sources and their scope are retained with the relevant entries.

Every claimed architectural benefit is conditional on its specified external contract. Quantitative latency, memory, implementation effort, upstream willingness, and patch-level independence have not been established. The packages are sufficient routes worth investigating, not proven minimal change sets. Some required properties could be supplied by an adequate Anneal adapter rather than a new upstream interface.

Distinguish historical observed limitations, hypothetical strengthened contracts, logical consequences of those contracts, and open engineering questions. A newer tool version may already implement part of an opportunity; recheck the relevant source and operation before proposing work. A corrected report reopens the affected motivation, while a different version does not retroactively refute an exact older observation.

No compiler/prover, production-client, or resource experiment was rerun for this filing. Public documentation links from the review were checked for their narrowly cited facts; that is not an implementation capability audit or independent replication.


Agent origin: ChatGPT, Anneal project conversation, October 1, 2026. Filed at Josh Liebow-Feeser's explicit request to preserve the preceding upstream-capability review. No source-native conversation permalink or identifier was exposed. This footer records authorship and publication context, not technical evidence or adoption.

Dominant language
Rust
Stars
2.6k
Forks
194
Avg merge
18h 15m
Merged PRs (30d)
14

Getting set up

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

More from google/zerocopy

All issues in google/zerocopy

Similar issues

More Rust issues

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.